Malnutrition and dehydration are key problems in the elderly population. The causes for these problems are that seniors frequently skip meals and forget to drink water. As medication is often taken with meals, skipping a meal usually results in missing to take the medication and this leads to poorer medication compliance.
By recognizing eating and drinking gestures during the day, it is possible to trigger smart reminders that will advise the user to take a medication in the key moment when he/she is eating, or alert that he/she is skipping a meal, or forgetting to drink water for a long time.
This gesture recognition may be based on analysing the signals of inertial sensors from off-the-shelf smartwatches, or other sensors used around the wrist, to observe and classify the movements performed during the day.
Since the arm has a great number of degrees of freedom and wrist movements and gestures are very diverse, the inertial sensor signals obtained from free daily living conditions in this position are expected to be very noisy. Therefore gesture recognition problems are not trivial and should be addressed using classifiers that are able to take into account both sequences of sub-gesture motions and model inter-gesture sequential dependencies, for example higher order Hidden Markov Models. In order to achieve high recognition accuracy, other levels of information, such as location and time of the day could also be added in a higher level classification layer.
Author: Diana Gomes
Type: MSc thesis
Partner: Faculdade de Engenharia da Universidade do Porto